Integrated Deep Learning With a Virtual Assistant Based on the Chatbot “Kak Sarah” in the Subject of History

Abstract

This paper presents the development and implementation of an AI-powered virtual assistant-based chatbot system called “Kak Sarah” designed to enhance deep learning in history education. The system integrates Retrieval-Augmented Generation (RAG) algorithm with OpenAI GPT-5 API, utilizing a Laravel backend with Docker containerization and React frontend to create an adaptive learning management system. The chatbot employs supervised fine-tuning techniques to enable role-playing simulations where the AI assumes the persona of historical figures, facilitating Socratic questioning mechanisms to promote critical thinking. The system architecture incorporates ElevenLabs text-to-speech technology for accessibility compliance and implements offline-first design principles for improved performance. Implementation across five schools involved 312 students, with performance evaluation conducted through paired and independent sample t-tests. Results demonstrated significant learning outcomes with an average score improvement of 24.35 points between experimental (85.94) and control classes (61.59), achieving statistical significance (p = 0.005). The platform successfully implements adaptive learning algorithms that adjust content difficulty based on student cognitive levels. Additionally, teacher computational competency assessments showed improvement from 3.67 to 4.75 on a 5-point scale. This study validates the effectiveness of RAG-based conversational AI integrated with pedagogically-structured dialogue frameworks as a viable approach for technology-enhanced learning systems.